Advances in Aircraft and Spacecraft Science, Vol. 4, No.

6 (2017) 613–637
DOI: https://doi.org/10.12989/aas.2017.4.6.613 613

Damage detection through structural intensity and vibration

based techniques
G. Petrone* , A. Carzana, F. Ricci and S. De Rosa
Department of Industrial Engineering, Aerospace Section, University Federico II, Via Claudio 21, 80125
Naples, Italy

(Received July 27, 2017, Revised July 30, 2017, Accepted July 31, 2017)

Abstract. The development systems for the Structural Health Monitoring has attracted considerable
interest from several engineering fields during the last decades and more specifically in the aerospace
one. In fact, the introduction of those systems could allow the transition of the maintenance strategy from
a scheduled basis to a condition-based approach providing cost benefits for the companies. The research
presented in this paper consists of a definition and next comparison of four methods applied to numerical
measurements for the extraction of damage features. The first method is based on the determination
of the Structural Intensity field at the on-resonance condition in order to acquire information about the
dissipation of vibrational energy throughout the structure. The Damage Quantification Indicator and the
Average Integrated Global Amplitude Criterion methods need the evaluation of the Frequency Response
Function for a healthy plate and a damaged one. The main difference between these two parameters is
their mathematical definition and therefore the accuracy of the scalar values provided as output. The
fourth and last method is based on the Mode-shape Curvature, a FRF-based technique which requires the
application of particular finite-difference schemes for the derivation of the curvature of the plate. All the
methods have been assessed for several damage conditions (the shape, the extension and the intensity of
the damage) on two test plates: an isotropic (steel) plate and a 4-plies composite plate.
Keywords: damage detection; structural intensity; curvature method

1. Introduction
Accumulation of damage among structure can cause severe structural failures. Development of an
early damage detection method for structural failure is one of the most important keys in maintaining
the integrity and safety of structures.
Many different damages can be modelled depending on the cause of the damage, on the sizes of
the involved elements and on the material of the assessed structure.
In the work by Liu and Swaddiwudhipong (1997), damages derived by low-velocity impacts have
been modeled through the definition of degenerated shell elements that have been deformed by the
contact-impact (further study by El-Abbasi and Meguid (1998), Liu et al. (2004)). However, this

*Corresponding author, Ph.D., E-mail: [email protected]

Copyright © 2017 Techno-Press. Ltd.

https://www.techno-press.org/?journal=aas&subpage=7 ISSN: 2287-528X(Print), 2287-5271(Online)
614 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

type of damage only shows the effect of the impact on the shape of the elements and then on the
behaviour of the deformed structure, whereas none of the material properties have been modified by
this damage modelling.
An improved modeling technique for damages has been developed recently, focusing on the as-
sessment of nonlinear features of damages. The basic model of a nonlinear crack using finite elements
is to have disconnected elements with collocated nodes, as used by Lee et al. (2006). This scheme,
although easy to implement, can lead to non-physical conditions as the crack surfaces do not inter-
act and structural elements may end up overlapping. Accordingly, this approach cannot take into
account for contact friction or interaction forces between the crack surfaces. Therefore, this scheme
should only be used when the physical crack or damage feature is actually behaving as disconnected
elements with negligible interaction.
An improved model of a crack is the bilinear stiffness model assessed deeply by Hambric et al.
(2012) (also by Al-Shudeifat and Butcher (2011) for cracked beams) which accounts for changes in
stiffness at a node over time. As the stiffness between the surfaces changes depending on whether
the crack is “open” or “closed”, the stiffness alternates between two values assigned to the open and
closed states. However also this method requires accurate contact modelling in order to obtain a
realistic change of the dynamic behaviour of the plate due to the presence of the crack.
Many different methods for the detection of such a damage have arisen during the last decades,
based on the determination of the change in structural response due to the reduction of material elastic
properties. Structural intensity (SI) reveals the magnitude and direction of mechanical energy flow,
or power flow, through a structure. Knowledge of SI fields is vital in the study of structure-borne
noise as it reveals the sources and sinks of vibrational energy, as well as the dominant paths of energy
flow, as functions of both space and frequency. The vibrational energy features can be very useful
for general monitoring purposes and in particular can be of great importance for damage detection.
The SI field determination is possible through the measurement of internal forces and displacement,
hence it can be measured experimentally by using accelerometers for flexural wave propagation in
beam and plate structures (first measurements were obtained by Noiseux (1970)).

Fig. 1 FRF of a damaged plate

 Damage detection through tructural intensity and vibration based techniques 615

All the SI measurements methods require large numbers of sensors in order to obtain detailed
power flow maps. To complement the experimental SI methods, numerical methods have been de-
veloped since they allow for mapping the intensity over the entire structures; this is particularly useful
when experimental techniques are extremely limited or not possible. Moreover, benefits of numerical
analysis include no equipment requirements or experimental loading and ease of adjusting loading
and drive conditions. Numerical-based SI methods have also been developed primarily using finite
element analysis (FEA), indeed it is cost-effective for low- to mid-frequency analysis and meshes are
relatively easy to create and to analyse using commercial software packages. The SI field is able to
show the mechanical energy paths in a clearer way whether the structure is assessed at on-resonance
condition. Indeed if the forcing frequency is close to one of the natural frequencies, then the system
will show a resonance response and hence greater vibrations. The onset of vibration waves is related
to the mechanical energy flow, and this is the reason why the SI based methods seem to be the most
valuable techniques for structure response analysis under resonance condition.
Similar considerations has led to the definition of other damage detection methods based on the
evaluation of the frequency response function (FRF) under resonance condition. Unlike the SI based
techniques, these methods require no modal analysis or other complex post-processing operations
to obtain an estimate of relative structural health. As it is possible to see from Fig. 1, the FRF of a
plate slightly changes with the increase of the damage extension, obtained reducing the mechanical
properties of a given area. Hence it is possible to compare the frequency response of a damaged plate
with the one of a healthy plate. It is easy to see that the higher distance between the curves is obtained
at the natural frequencies, i.e., close to the resonance peaks, and therefore it has been demonstrated
why the assessed FRF based techniques need the evaluation of the response function at on-resonance
condition.
Although their ease of use, the damage detection methods based on the determination of the sys-
tem response (squared acceleration of all nodes of the model) to a fixed excitation force are not able
to track the position of the damaged area, but only its presence and extension. Therefore another FRF
method has been applied in order to provide information about the position and intensity of the dam-
age. The analysis of the mode shapes curvature makes possible to detect the crack location through
the assessment of the change of the material properties, and in particular of the elastic modulus. The
derivation of the curvature is possible through the application of various finite difference schemes to
the nodes displacement value, hence yielding an accuracy of the method that depends on the density
of the mesh.
Recently, two interesting papers appeared in literature, Jalali and Noohi (2018) and Guo and Xu
(2018): they certified the active interests in engineering research field for such applications.
In this paper the background on structural intensity is given in Section 2 together with the detection
methods to be investigated. Section 3 presents a summary of the numerical models and the results
are in Section 4. Some concluding remarks in Section 5, close the work.

2. Theoretical background and models

2.1 Structural Intensity
The “vibration intensity”, or “structural intensity” (the denomination comes from the analogy with
acoustic intensity), gives significant information about the strength and location of sources, sinks and
616 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

transmission paths of structure energy through the plotting of a vector map. The methods based on
the analysis of the SI field hence provide relevant datas about the structure vibro-acoustic properties
that can be used to better understand the behavior of the studied model (e.g. a plate). Although this
technique is applicable to every type of vibration energy wave, most of the focus during the years has
been put on the flow of flexural waves. Indeed this type of waves are the most relevant in order to
analyse the energy radiation and dissipation of the structure.
The determination of the Power Flow or the Structural Intensity in simplified structures such as
thin plates has been the focus of much work in recent years (Verheij (1980), Cuschieri (1987), Arruda
and Mas (1996), Hambric (2009), Petrone et al. (2016)). The difference between all the methods
developed by several authors during the last about 40 years is represented by the procedure for the
input measurements. These measurements can be accelerations, displacements, rotational displace-
ments, etc. The definition of Instantaneous Power Flow is due to Noiseux (1970), and represents an
energy flux, a power rather than a flow intensity. For the case of a two dimensional structure such as
a plate, it is possible to express the components of this vector in the direction x and y as follow

Px (x, y) = Qx ẇ − Mxy θ̇x + Mx θ̇y (1a)

Py (x, y) = Qy ẇ − Mxy θ̇y + Mx θ̇x (1b)

Typically, the Time Average Power Flow is more easily measurable and then of major interest with
respect to the instantaneous power flow, hence the Eq. (1) can be easily modified in

⟨ ⟩ ⟨ ⟩
Px (x, y) = ⟨Qx ẇ⟩T − Mxy θ̇x + Mx θ̇y (2a)
⟨ ⟩T ⟨ ⟩T
Py (x, y) = ⟨Qy ẇ⟩T − Mxy θ̇y + My θ̇x (2b)
T T

where: Qx and Qy are respectively the shear forces in the x and y directions, Mx and My are the x
and y directed bending moments, Mxy is the twisting moment, ẇ is the normal velocity, θ̇x = ∂w
∂y and

∂x are the time rate of change of slope in the x and y directions, ⟨ ⟩T denotes the time average.
θ̇y = ∂w
Few years after Noiseux’ studies, Pavic (1976), proposed a novel concept for power flow in struc-
tures, called Structural Surface Intensity, or simply Structural Intensity (SI), no more based on mea-
surement of forces and moments acting on the plate, but on the evaluation of internal normal and
shear stress. If we consider that σx and σy are the surface normal stress components, τxy is the sur-
face shear stress, ẇx and ẇy are the surface velocities, and ⟨ ⟩T denotes the time average, then the
definition of SI components based on Pavic works can be expressed as follows

Ix = − ⟨σx ẇx ⟩T − ⟨τxy ẇy ⟩T (3a)

Iy = − ⟨σy ẇy ⟩T − ⟨τxy ẇx ⟩T (3b)

The determination of the internal forces-displacements or stress-strain at all the nodes of the model
can be obtained through experimental or numerical methods. The experimental methods are based
on the application of several linear and rotational accelerometers to the surface of the plate. Their
accuracy is highly dependent on the number of measurements and on the specific finite difference
 Damage detection through tructural intensity and vibration based techniques 617

scheme applied. Moreover the experimental measurements show several errors due to the inaccuracy
of the instruments and the attachment method. In numerical investigation, the acquisition of the forces
and moments (Power Flow assessment), or the stresses and strains (Structural Intensity) are carried
out by a FE model analysis, as shown by Gavric and Pavic (1993). The Eq. (3) can be generalized
by considering the components of the SI in the ith and j th directions. Then by neglecting the effect
of the shear stress, the SI component can be defined as follows
Ii = Ii (t) = −σij (t)ẇij (t) i, j = 1, 2, 3 (4)
where σij (t) and ẇij(t) are the stress and velocity in the j th -direction at time t. It is possible to
replace the velocities with the displacements by using the commonly adopted complex algebra and
since the stresses and displacements are usually determined through an integration over the thickness,
then the SI can be expressed as a power flow per unit width. The expression for the two components
of the SI can be hence expressed as
ω
Ix = − I[Nx ũ + Nxy ṽ + Qx w̃ + Mx θ̃y − Mxy θ̃x ] (5a)
2
ω
Iy = − I[Ny ṽ + Nxy ũ + Qy w̃ − My θ̃x + Mxy θ̃y ] (5b)
2
where Nx and Ny are the complex axial forces, Nxy is the complex in-plane shear force, Qx and
Qy are the complex transverse shear forces, Mx , My and Mxy are the complex bending moments,
ũ, ṽ and w̃ are the complex conjugate of translational displacements, θ̃x , θ̃y and θ̃z are the complex
conjugate of rotational displacement along x, y and z directions. I denotes the imaginary part.
This method has become widely used because of its very low cost and its ease of execution, and
also for its substantial reduction of experimental errors. In addiction to this, the coupling between a
numerical and a non-contact method, for example a laser acquisition, may allow a very fast and accu-
rate real-time control of plates and panels. On the other hand, great attention must be put on the choice
of the FEM elements (e.g., beam, shell or plate elements) in order to obtain a good approximation of
the experimental model.
Through finite element simulation, Lim et al. (2006) have shown that damage causes fundamental
changes in the structural intensity fields of plates. In these analytical studies, changes in magnitude
and phase angle of the intensity fields were observed in areas directly surrounding the crack. When
a structure presents any type of crack damage, energy flow must divert around the crack, causing
local changes in SI vector divergence. When determination of the Structural Surface Intensity over
an entire structure is possible, it shows potential for use in damage localization and indication of
damage severity. However when determination of the SI field over an entire structure is not possible
or too difficult, classical FRF techniques can be used to show the presence of a damaged area.
2.2 The detection methods
The damage detection methods that will be here presented, are based on the assessment of the Fre-
quency Response Function defined as the ratio between the squared acceleration of the nodes (output)
and the fixed excitation force (input). Hence these methods can be applied to directly measured fre-
quency response data, providing a real-time monitoring of the structural health.
A widely used metric for the damage detection in a mode shape is the Modal Assurance Criteria
(MAC). The MAC is used to find correlation characteristics between two given matrices: the first
618 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

matrix consists of the first N mode shapes of an healthy plate, and the second consists of the same N
mode shapes in a damaged plate. The MAC is represented by a [N xN ] matrix that provides informa-
tion of the correlation between the two sets of different mode shapes, and therefore where a lack of
correlation denotes the presence of a damage. Since the MAC has proven helpful with applications
in damage detection through use of mode shape data it has been expanded into the frequency domain.
A direct expansion of the MAC into the frequency domain yields the Response Vector Assurance
Criterion (RVAC) and the Global Amplitude Criterion (GAC).
In the MAC, each mode shape in a system reference state is compared to each mode shape in a
current state. This provides a resultant matrix where the diagonal terms represent comparison be-
tween same mode shapes. Hence being the RVAC a direct expansion of the MAC into the frequency
domain, it is only applied along the equivalent of the diagonal of the MAC matrix. This means that
only similar frequencies (same modes) are compared between the damaged and undamaged states.
The RVAC, is defined as
∑n d 2
i=1 αi (ω)αi (ω)
∗
RV AC(ω) = ∑n [ d d∗
] ∑n (6)
i=1 αi (ω)αi (ω) i=1 [αi (ω)αi (ω)]
∗

where α(ω) is the healthy FRF at frequency ω, αd (ω) is the damaged FRF at frequency ω, and ∗
denotes the complex conjugate.
The input for the RVAC at a given frequency is a set of two vectors. One vector contains the
magnitude of the FRF at each degree of freedom for a reference case, and the other vector contains
equivalent quantities for a damaged or current case. The output of the RVAC for a single frequency
is a scalar quantity between 0 and 1, where a RVAC value of 1 corresponds to perfect correlation
and 0 corresponds to no correlation. The RVAC can be applied at any number of frequencies (in this
study just the first 9 modes have been considered) such that the total output of the RVAC is a vector
containing one scalar value for each frequency evaluated.
Another frequency response-based metric that determines correlation between a baseline and cur-
rent system state is the GAC. Similar to the RVAC, the GAC is based on a 0 to 1 scale where 1
corresponds to perfect correlation and 0 corresponds to no correlation.
The GAC takes the form defined as
n
1∑ 2 αid∗ (ω)αid (ω)
GAC(ω) = (7)
[αi∗ (ω)αi (ω)] + αid∗ (ω)αid (ω)
[ ]
n
i=1

The GAC uses the same input as the RVAC (two vectors containing frequency response data for
a healthy and current structural state), and the output is a 1 x n vector where n represents the number
of frequencies at which the reference and current states are compared. While on the surface the
GAC may look very similar to the RVAC, the mathematical derivation of these two parameters are
slightly different. Nevertheless, the output of the GAC is a vector with a length equal to the number
of frequencies studied, which is comparable to the RVAC.
It is not always helpful to have to interpret data in vector form. In reality, it is much simpler to
handle a single number that may become an indicator of damage when it crosses a certain threshold.
As a result, the Damage Quantification Indicator (DRQ) and the Average Integrated Global Ampli-
tude Criterion (AIGAC) were developed. The DRQ converts the RVAC vector output into a scalar
 Damage detection through tructural intensity and vibration based techniques 619

quantity by averaging all of the values of the RVAC. In practice, the DRQ can be assigned a critical
threshold. If the DRQ falls below this value, critical structural damage has been achieved, and repairs
to the structure have to be undertaken. The DRQ is defined as
Nω
1 ∑
DRQ = RV AC(ωk ) (8)
Nω
k=1

where Nω represents the total number of frequencies for which the RVAC is calculated. The DRQ
shows potential in damage detection applications because it is an FRF based damage detection method
where the output is a single number. This implies that it can be used in real-time systems where a
“damaged” or “undamaged” indicator can be used. One shortcoming of the DRQ is that it cannot be
used as a damage localizer as it is a global damage detection technique. This is simply a consequence
of the DRQ’s scalar output. The AIGAC has the same definition of the DRQ, but it represents the
average value of the GAC vector. The AIGAC is defined as
Nω
1 ∑
AIGAC = GAC(ωk ) (9)
Nω
k=1

2.3 Mode shape curvature method

The Mode Shape Curvature method has been used here then to provide complementary informa-
tion with respect to the previously cited FRF methods. To implement this method, measured nodes
displacement value at all experimental grid points are isolated at specific frequencies (plate natural
frequencies). By isolating the transfer functions at a single frequency, an operational “mode shape”
at the frequency of interest is obtained. The presence of a damage is the cause for a reduction of
elastic modulus into an area, but this has an effect on the curvature of a plate through the definition
of the flexural stiffness
( 2
∂2w
)
∂ w
Mx = −D +ν 2 (10a)
∂x2 ∂y
( 2
∂2w
)
∂ w
My = −D + ν (10b)
∂y 2 ∂x2

where
Eh3
D= with: h = plate thickness (11)
12(1 − ν 2 )
The curvature of the plate, i.e., the 2nd -order derivative of the displacement in the in-plane direc-
tions, can be therefore studied in order to acquire information about the presence of a damage. The
approximated derivatives shown in Eq. (10) have been obtained through the application of a cen-
tral finite difference scheme (curvature in border elements has been derived through f orward and
backward schemes instead).
A further study has been performed in order to determine the most accurate definition of curvature
for the damage detection. Two types of curvature have been defined and assessed
620 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

1 ∂2w ∂2w
( )
Mean Curvature : κmean = + (12a)
2 ∂x2 ∂y 2
∂2w ∂2w
Gauss Curvature : κGauss = · (12b)
∂x2 ∂y 2
The evaluation of the curvature values of the elements can be derived by averaging the curvature
values of the four corner nodes of each square element. The elements curvature has to be found for
all the natural frequencies included in an arbitrary range (in this study the first nine have been taken),
and then it is possible to calculate for each element the residual value through the formula expressed
in Eq. (13) for the ith element
Nf
(13)
∑
Ri = κdi,n − κhi,n
n=1

where Nf is the number of natural frequencies considered, D and H denote that the curvature is
related to the damaged or the non-damaged (healthy) plate respectively.
The plotting of the residuals is used to show clearly the effect of the presence of a damaged area
on the modes shape and hence on the modes curvature.
2.4 Summary
The four detection methods to be tested inside this work are thus

• RVAC
• GAC
• DRQ
• AIGAC

Given the nature of the AIGAC calculation, it has the same inherent advantages and disadvantages
in damage detection applications as the DRQ. The trait that separates the DRQ from the AIGAC is
the formulation of the GAC. The AIGAC is an averaged integration of the GAC, a lesser used and
lesser known formulation than the MAC. Since the GAC is not as well known in damage detection
applications as the MAC, the true potential of the AIGAC in damage detection is unknown.
Both the DRQ and the AIGAC can be used as global damage detection techniques, and hence are
not able to determine the position and shape of the damaged area or crack.

3. Numerical models
3.1 Damage
In this paper, the damage studied has been modeled as a fatigue crack that involves just some
elements of the plate (Simmermacher et al. (2012)).
 Damage detection through tructural intensity and vibration based techniques 621

Fig. 2 Plate model

For a finite element model with N elements, the elastic moduli for the elements can be denotes as

E = {E1 , E2 , . . . , Ei , Ei+1 , . . . , Ei+n , . . . , EN −1 , EN } (14)

The damage is represented by a change in the flexural rigidity or stiffness of the plate, which, in
turn, is reflected by a reduction in the elastic modulus of the damaged elements. If the model contains
n damaged elements, their elastic moduli {Ei+1 , . . . , Ei+j , . . . , Ei+n } are replaced by

Ed = {ki+1 Ei+1 , . . . , ki+j Ei+j , . . . , ki+n Ei+n } (15)

where ki+j , j = 1, . . . , n, denotes the reduction factor (lower than 1) for the (i + j)th damaged
element.
3.2 Plates
The assessed structure is a rectangular plate with size 1080×1440 mm, Fig. 2, and is modelled
through a mesh of 18×24 four nodes elements. The plate is simply supported on all edges and is
subjected to a nodal excitation force applied at position P(300 mm, 1260 mm). The applied force has
a magnitude of 1N and a frequency of 20 Hz, indeed it is possible to assess through a normal mode
analysis that 20 Hz is a value close to the first natural frequency. The mechanical energy introduced
into the plate is dissipated by a viscous damper attached at the position Q(780 mm,180 mm), defined
by a damping coefficient ζ = 0.5. Both the force and damper locations are represented in Fig. 2 by
yellow vertical arrows indicating the selected node.
Two different materials have been chosen for the analysis performed in this dissertation in order
to provide a complete set of information about the applicability of these damage detection methods
to different structures. The first assessed plate has a thickness of 6 mm and is made of isotropic steel
having the properties shown in Table 1.

Table 1 Isotropic steel properties

E G ν ρ
205.6 GP a 79.15 GP a 0.3 7800 kg/m3
622 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

Table 2 Orthotropic lamina properties

Ex Ey Gxy Gxz Gyz νxy νxz νyz ρ
175.1 GP a 6.7 GP a 4.1 GP a 2.056 GP a 4.1 GP a 0.254 0.469 0.469 1520 kg/m3

The second study case is about a composite plate made of four orthotropic laminas stacked together
and having the following orientations: ±90◦ / 0◦ / 0◦ / ± 90◦ . The single lamina is a 2 mm thick
composite ply made of graphite fibers and epoxy resin matrix, therefore it has been given the material
properties exhibited in Table 2.
Note that the x-direction is parallel to the shorter side of the plate, and the y-direction is then
perpendicular and directed as the long edge. The structural damping parameter η has been set equal
to 0.007 in both study cases. This value is consistent with the measurements made on sample metallic
and composite structures. A higher structural damping would produce a greater dissipation of the
mechanical energy flow along its path, and therefore a decrease of the damper effectiveness.
3.3 Modal information
Numerical simulations were conducted, in terms of modal analysis and frequency response func-
tion, using the commercial finite element solver Nx/Nastran, focusing on the low frequency region.
The first natural frequencies of both investigated panels, under simply-supported boundary condition,
are reported in Table 3.
In order to simulate a structural fault, the Young modulus of some elements of the isotropic plate
model, has been reduced progressively: the results of [10, 30, 50, 70]% variations have been obtained.
All analysis have been carried out by starting from a damaged area of 120 mm and extending this up
to 480 mm. Results obtained by these analysis are represented in Table 4. It can be noticed that the
frequencies shift methods are not very sensitive to bending stiffness variation. As a matter of fact,
with a big reduction of the Young modulus (70%) in a region, the variation of the value of the modal
frequencies is too small to be detected and hence a different approach is needed.

Table 3 First natural frequencies of Isotropic and orthotropic panels

Mode Steel plate Graphite-epoxy plate
hline number Natural frequency [Hz] Natural frequency [Hz]
1 19.73 31.92
2 40.94 36.88
3 57.60 48.80
4 76.32 69.01
5 78.26 97.21
6 112.78 123.75
7 120.72 126.70
8 125.86 132.88
 Damage detection through tructural intensity and vibration based techniques 623

Table 4 First natural frequencies of isotropic plate for different damage intensity and dimension
Damage extension E modulus reduction
10% 30% 50% 70%
Mode 1 [Hz] 19.72 19.71 19.64 19.66
Mode 2 [Hz] 40.94 40.94 40.93 40.93
120mm
Mode 3 [Hz] 57.60 57.60 57.59 57.59
Mode 4 [Hz] 76.29 76.22 75.91 76.02
Mode 1 [Hz] 19.71 19.64 19.56 19.43
Mode 2 [Hz] 40.93 40.93 40.92 40.91
240mm
Mode 3 [Hz] 57.59 57.55 57.51 57.45
Mode 4 [Hz] 76.23 75.91 75.47 74.73
Mode 1 [Hz] 19.69 19.61 19.49 19.30
Mode 2 [Hz] 40.93 40.92 40.91 40.88
360mm
Mode 3 [Hz] 57.56 57.45 57.34 57.18
Mode 4 [Hz] 76.15 75.72 75.06 73.92
Mode 1 [Hz] 19.68 19.58 19.43 19.18
Mode 2 [Hz] 40.93 40.91 40.88 40.85
480mm
Mode 3 [Hz] 57.52 57.33 57.09 56.77
Mode 4 [Hz] 76.11 75.57 74.74 73.25

(a) Healthy Composite Plate: Horizontal fibres (b) Healthy Composite Plate: Vertical fibres

Fig. 3 Graphite/Epoxy Composite Plate, SI method for different fibers orientation

624 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

(a) Damage extension: 120 mm (b) Damage extension: 240 mm

(c) Damage extension: 360 mm (d) Damage extension: 480 mm

Fig. 4 Steel plate, SI method for different crack extension

 Damage detection through tructural intensity and vibration based techniques 625

4. Damage detection
4.1 Isotropic plate
The Structural Intensity based method has been applied to the isotropic steel plate in order to detect
the presence of a damage, and in particular to assess its size and intensity.
Five cases have been examined corresponding to 5 different lengths of the central damaged area:
• healthy plate;
• 120 mm (11.1% of plate width);
• 240 mm (22.2% of plate width);
• 360 mm (33.3% of plate width);
• 480 mm (44.4% of plate width);
In this first study the damage intensity has been fixed to 70%; this means that the elastic properties
(E and G) of the elements belonging to the damaged area have been decreased of the 70% of their
original value. As expected, the SI field of the damaged plates represented in Fig. 4 shows the path
of mechanical energy that flows from the source of vibrations, i.e., the external force position at the
top-left corner, to the viscous damper at the bottom-right corner of the plate.
The reduction of elastic stiffness due to the presence of a damage has a deep effect on the diver-
gence of the SI flow, hence producing a change in the vectors phase and magnitude for the elements
belonging to the damaged region or close to it. Moreover it can be seen through the scale on the right
side of the images that the presence of a crack has a relevant effect also on the global magnitude of
SI, i.e., on the average value of |SI|. A further analysis on the steel plate SI field has been developed
in order to provide information about the effect of the increase of damage intensity on the flow of
vibrational energy. In this study case the damage extension has been fixed to a length of 240 mm,
i.e., about the 22.2% of the plate width, while the damage intensity varies in a range that goes from
30% to 90%.
Even in this case, as can be seen from the Fig. 5, the damage intensity has an effect on the global
magnitude of structural intensity that hence is reflected by a variation of the scale on the right of the
pictures. The increase of damage intensity yields to a significant deepening of the “hole” representing
the crack, and therefore a more accentuate deviation of the SI flow.
The SI field of the plate has been plotted by using a specific MatLab function (imagesc) that
automatically defines the scale represented by the color bar beside the figures. Therefore it is easy to
see that different crack conditions (damage extension and intensity) determine a particular magnitude
color scale, then making difficult to compare the figures in order to make proper conclusions. This
problem has been solved through the definition of a new parameter that could make the comparison
between different crack conditions easier, and then representing a more effective damage detection
method.
A “reference” value of the SI magnitude has been derived for each element of the horizontal band
passing through the damaged area. Hence this band has a width equal to the plate size in the x-
direction, i.e., 1080 mm, and a height of 60 mm. The reference value of SI magnitude for the ith
element (∆SIi ) has been derived through the Eq. (16), with i = 1, 2, ...Nb and Nb that is the number
of elements belonging to the selected horizontal band.
626 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

(a) Damage intensity: 30% (b) Damage intensity: 50%

(c) Damage intensity: 70% (d) Damage intensity: 90%

Fig. 5 Steel plate, SI method for increasing damage intensity

 Damage detection through tructural intensity and vibration based techniques 627

Fig. 6 ”Reference” Structural Intensity for central band elements: effect of crack extension

MiD MiH
∆SIi = D
− H
(16)
Mav Mav
where MiD is the SI magnitude of the ith element in the damaged plate, Mav D is the SI average magni-

tude of the ith element in the damaged plate, MiH is the SI magnitude of the ith element in the healthy
plate, and MavH is the SI average magnitude of the ith element in the healthy plate.

This modified definition for the SI can be applied in order to derive a more accurate SI based
damage detection method. The SI flow diverts from its original path (the healthy plate case) then
yielding an increase of the magnitude in the elements directly surrounding the crack. This is reflected
in Fig. 6 as positive peaks of ∆SI outside the range delimited by the two vertical dotted lines (the

Fig. 7 ”Reference” Structural Intensity for central band elements: effect of damage intensity
628 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

(a) Damage Intensity < 50%

(b) Damage Intensity ≥ 50%

Fig. 8 DRQ and AIGAC plot for isotropic steel plate

 Damage detection through tructural intensity and vibration based techniques 629

crack position). On the opposite, the damaged elements, i.e., the ones that are inside the cited range,
show a drop of ∆SI that is more intense with the decrease of the crack extension.
It can be noted that the crack tip location corresponds to the external points where there occurs the
value of ∆SI goes from positive to negative, or the opposite. When the damage extension is high,
the internal elements of the crack, i.e., far from the crack tips, show an increase in magnitude due to
the fact that the flow is not able to provide an adequate deviation, and hence is forced to pass through
the crack width.
The study of the ∆SI for several damage intensities shows that through this method it is possible to
detect even very soft damages, as it is exhibited in Fig. 7. Indeed for very small cracks the targeting of
the damaged region is eased with respect to the case of great crack extension because all the damaged
elements exhibit a negative value of ∆SI whereas all the other external elements show positive
values.
Despite their accuracy, the SI based damage detection methods are not so common because of the
great effort required for the measurements. Therefore, FRF based techniques are preferable in some
cases. The assessment of DRQ and AIGAC has been carried out for the same damage conditions
already studied before, i.e., the healthy plate case and four cases for an increasing crack extension.
By looking at Fig. 8 it is possible to make some considerations about these two parameters used to
investigate the presence of a damaged area inside the plate.
The sensitivity of these methods has been obtained, indeed both DRQ and AIGAC do not seem
to be able to detect any damage if the intensity is too low, e.g. less than the 50%. In this case their
values is about constant and equal to 1 for every extension of the crack, and this means that the FRFs
of the healthy and damaged plate are exactly the same, i.e., no significant damage is observed.
All the numerical simulations performed with damage intensity values above or equal to 50% (i.e.,
90%, 70% and 50%) have yielded exactly the same graph for DRQ and AIGAC, hence this method
is not able to determine the intensity of a damage, but just its existence and the crack length. Both the
DRQ and the AIGAC decrease with the increase of the crack extension, and this suggests a damage
maintenance criterion based on the definition of an arbitrary threshold (a sample value could be 0.92)
that indicates whether a crack maintenance intervention needs to be applied.
The DRQ curve is placed in a slightly lower position than the AIGAC curve, therefore showing
to be a more conservative approach and hence preferable with respect to the AIGAC.
The last FRF based damage detection technique that has been applied to the isotropic steel plate
is the mode shape curvature method. This is based on the derivation of the curvature along the x
and y directions from the measurement of the nodal displacements. Hence the input FRF vectors are
different with respect to the ones required for the application of the DRQ and AIGAC methods.
The difference of the mathematical definition between the mean and Gaussian curvature yields a
change in the damage detection power of the curvature method (Fig. 9). The assessed plate shows
a central damage with an extension of 240 mm and a very low intensity, i.e., a reduction of elastic
stiffness equal to the 30% of the value corresponding to a healthy plate.
The curvature method shows to be ineffective when the damage intensity is too low, indeed in
this case both the mean and the Gaussian curvature do not highlight clearly the shape of the crack.
It is possible to observe that the Gaussian curvature is less accurate than the mean curvature because
it leads to a wrong prediction of the damage position (a feasible crack location might be at the top-
right and bottom-left corners, based on Fig. 9(b)). However the accuracy of the mean curvature
630 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

(a) Mean Curvature

(b) Gauss Curvature

Fig. 9 Curvature method: damage extension of 240 mm and intensity of 30%

Damage detection through tructural intensity and vibration based techniques 631

(a) Mean Curvature

(b) Gauss Curvature

Fig. 10 Curvature method: damage extension of 480 mm and intensity of 70%

632 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

drastically decreases with the increase of the crack extension, making this method useless for the
damage detection.
Opposite considerations can be made when the same methods are applied to a plate that presents
a deeper and larger crack, e.g., 70% of damage intensity and 480 mm of crack extension, such as
the one shown in Fig. 10. Obviously an intense damage is more easily recognized by any damage
detection technique, however it is possible to note that the Gauss Curvature method is more accurate
than the Mean Curvature method. Indeed it is possible to track the right contour of the crack only by
looking at Fig. 10(b)), whereas Fig. 10(a)) provides a lower gradient between the damaged and the
undamaged area of the plate.
The different mathematical definition of the mean and Gaussian curvature makes impossible to
compare the magnitude of the residuals deriving from the applied method (the first is of the order
of 10−7 whereas the second is of the order of 10−13 ), but it determines a complementarity of the
information provided. In conclusion it is essential to assess both the curvature residuals field in order
to detect the presence of a crack in all damage conditions.
4.2 Composite plate
The study of an isotropic steel plate has provided relevant information about the power of the SI
and FRF based methods for damage detection. However the application of the same methods to a
composite plate can be used to make also some considerations on the role that the single plies have
on the transmission of vibrational energy throughout the plate.
The high level of anisotropy of the graphite/epoxy composite lamina produce a significant change
in the power flow path with respect to the case of an isotropic (e.g., steel) plate. Indeed the mechanical
energy is more easily transmitted through materials featuring high values of the elastic modulus (this
is also the reason why a damage that is modelled as a region with a lower elastic stiffness yields a
hole into the SI field). This means that the SI field completely changes with the change of the fibres
orientation, and in particular the SI flow vectors are longer (i.e., higher magnitude) when the fibres
have a direction that is about parallel to the dominant direction of the flow.
The comparison between a steel plate and two composite (Graphite/Epoxy) plates featuring or-
thogonal fibers is shown in Fig. 3. It can be easily seen how the application of vertical fibres to
the composite plate is the choice that produces the greatest transmission of energy throughout the
plate and hence the highest SI magnitude. On the opposite the choice of horizontally directed fibres
strongly reduces the flow between the force and the damper, therefore decreasing the effectiveness of
the viscous damper. The global magnitude shown by this last case is of the order of 10−6 , about one
thousand times smaller than the SI magnitude of the plate with horizontal fibers, and even a hundred
times smaller than the magnitude of the steel plate (see Figs. 4 or 5 as reference).
From the reported data and presented models, it is clear that a damage applied to the plies that are
more involved in the SI transmission will be then detected with a greater accuracy with respect to
damages applied to less important plies. The composite plate analysed here (and earlier presented)
has been modelled with the external lamina parallel to the plate long edge and the internal lamina in
the orthogonal direction, therefore it may be predicted that the damages applied to the external lamina
should be more easily detectable. This result have be obtained through the assessment of the SI field
for all crack conditions (Fig. 11). The comparison between the vibrational energy holes provided
by different cracks leads to the determination of the most severe damage, i.e., the one applied to the
 Damage detection through tructural intensity and vibration based techniques 633

(a) Damaged plies: 1st (b) Damaged plies: 1st and 4th

(c) Damaged plies: 2nd and 3rd (d) Damaged plies: 1st , 2nd and 3rd

Fig. 11 Composite Plate, SI method for different damaged plies

634 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

(a) DRQ plot

(b) AIGAC plot

Fig. 12 DRQ and AIGAC plot for a 4-plies composite plate

 Damage detection through tructural intensity and vibration based techniques 635

external plies as assumed before. However the global magnitude of the SI field that is highlighted by
the color scale on the right of the pictures, significantly varies with the change of the crack position
(the highest element SI magnitude goes from about 11 · 10−4 to 2.5 · 10−3 ), therefore a more accurate
method needs to be applied in order to make it possible a comparison between the several cases.
The graphs shown in Fig. 12 make possible a deeper understanding of what it has been exposed
previously. The study of the DRQ and AIGAC plot obtained through the application of FRF based
methods, show that the plies fibres orientation has a great effect not only on the SI field, but on the
FRF too. The damage has been applied to different plies in order to investigate their ‘importance”
for damage detection, and it has been found that the external plies (i.e., the 1st and the 4th ) are more
useful for this purpose. Indeed when the damage is applied to just 2 plies, i.e., we are assuming
a crack of the same intensity and volume, it is easy to see that both the DRQ and AIGAC curves
decrease with a higher gradient for that case with respect to the opposite condition of crack placed in
the 2nd and the 3rd plies.
Once again the DRQ proves to be a more accurate method than the AIGAC method, indeed the
curves in Fig. 12(a) exhibit a more constant slope with respect to the curves of Fig. 12(b). Moreover
the AIGAC curve for the case of a damage applied to the 1st , 2nd and 3rd plies (green line) shows
a too low value for small crack extension (11.1%) due to numerical errors that are implicit in this
method.
The damage intensity has been set to a value of 90%, indeed also for the study of a composite plate
both the FRF methods have shown a sensitivity that prevents the detection for too soft damages (below
50% of intensity). Obviously the chosen model for damages does not take into account the interaction
between near cracks, i.e., cracks applied to adjacent plies, hence assuming that each damaged volume
is independent yields an underestimation of the intensity of a crack and its effect on the transmission
of vibrational energy.

5. Conclusions
All numerical investigations that have been made in this work had the purpose of acquiring infor-
mation about the behaviour of the dynamic response of a rectangular cracked plate. Several studies
have been performed in order to assess the effects due to the variation of the crack extension and in-
tensity, material parameters and fibres orientation with respect to the power flow direction. The use
of FRF techniques such as the DRQ and the AIGAC has shown that the detection of an heavy damage
can be obtained through the comparison with a scalar threshold that does not have to be trespassed.
This limit is specific for the analysed structure, and moreover this method does have a sensitivity
that prevents the detection of not very intense damages. Anyway, the ease of use of this techniques
and the associated results encourage further investigations on the potentiality that have been shown
through this work. In particular the DRQ has proven to be a much more reliable tool for damage
detection than the AIGAC.
The localization of the damage has been obtained through the analysis of the SI field and through
the study of the Curvature (Mean and Gaussian) of the plate. These two methods have proven to
be both valid but characterised by different sensitivities to the damage extension and intensity. The
study of the SI flow has led to the conclusion that the reduction of elastic modulus in the damaged
area yields a hole in the SI field that can be easily observed. Once found the location of the damaged
636 G. Petrone, A. Carzana, F. Ricci and S. De Rosa

area, the crack features can be then derived through an assessment of the ∆SI magnitude of only
some selected plate elements. On the other hand the Curvature method has proven to be helpful on
a wide range of damage conditions, and provide accurate indication of the crack location. Globally
the FRF Curvature Method seems to be easier and more accurate than the SI based method.
In the last part of this paper, the study of the damages in a composite plate have shown important
results, indeed it has been demonstrated that the damage applied to single plies is detectable by the
FRF techniques that hence can clearly show which is the worst damage that a composite can withstand
without falling below the expressed threshold.

References
Liu, Z.S. and Swaddiwudhipong, S. (1997), “Response of plate and shell structures due to low velocity impact”,
J. Eng. Mech., 123(12), 1230-1237.
Shepherd, M.R., Conlon, S.C., Semperlotti, F. and Hambric, S.A. (2012), “Structural intensity modeling and
simulations for damage detection”, J. Vibr. Acoust., 134(5), 051004.
Lee, H.P., Lim, S.P. and Khun, M.S. (2006), “Diversion of energy flow near crack tips of a vibrating plate using
the structural intensity techniques”, J. Sound Vibr., 296(3), 602-622.
Noiseux, D.U. (1970), “Measurement of power flow in uniform beams and plates”, J. Acoust. Soc. Am., 47(1B),
238-247.
Pavic, G. (1976), “Measurement of structure borne wave intensity”, J. Sound Vibr., 49(2), 221-230.
Lim, S.P., Lee, H.P. and Khun, M.S. (2006), “Diversion of energy flow near crack tips of a vibrating plate using
the structural intensity technique”, J. Sound Vibr., 296(3), 602-622.
El-Abbasi, N. and Meguid, S.A. (1998), “Large deformation analysis of contact in degenerate shell elements”,
J. Numer. Meth. Eng., 43(6), 1127-1141.
Liu, Z.S., Lee, H.P. and Lu, C. (2004), “Structural intensity study of plates under low-velocity impact”, J. Imp.
Eng., 31(8), 957-975.
Al-Shudeifat, M.A. and Butcher, E.A. (2011), “On the dynamics of a beam with switching crack and damaged
boundaries”, J. Vibr. Contr., 19(1), 30-46.
Simmermacher, T., Cogan, S., Horta, L.G. and Barthorpe, R. (2012), Topics in Model Validation and Uncer-
tainty Quantification, Springer.
Verheij, J. (1980), “Cross-spectral density methods for measuring structure-borne power flow on beams and
pipes”, J. Sound Vibr., 70(1), 133-138.
Cuschieri, J. (1987), “Power flow as a complement to statistical energy analysis and finite element analysis”,
Stat. Energy Analy.
Arruda, J.R.F. and Mas, P. (1996), “Predicting and measuring flexural power flow in plates”, Proceedings of
the SPIE the International Society for Optical Engineering.
Hambric, S.A. (2009), “Power flow and mechanical intensity calculations in structural finite element analysis”,
J. Vibr. Acoust., 112(4), 542-549.
Petrone, G., De Vendittis, M., De Rosa, S. and Franco, F. (2016), “Numerical and experimental investigations
on structural intensity in plates”, Compos. Struct., 140 94-105.
Gavric, L. and Pavic, G. (1993), “A finite element method for computation of structural intensity by the normal
mode approach”, J. Sound Vibr., 164(1), 29-43.
Jalali, H. and Noohi, F. (2018), “A modal-energy based equivalent lumped model for open cracks”, Mech. Syst.
 Damage detection through tructural intensity and vibration based techniques 637

Sign. Proc., 98, 50-62.

Guo, T. and Xu, Z. (2018), “Data fusion of multi-scale representations for structural damage detection”, Mech.
Syst. Sign. Proc. 98, 1020–1033.

EC